Product

AquaBrain

Problem

Thai aquaculture is exposed to monsoon rain, salinity shocks, algal crashes, disease, feed waste, and strict export paperwork. Most sites still run on experience and lagging lab tests, so risk is found late.

Solution

Pond sensors and cameras feed an edge node for real-time inference. A central NVIDIA A100 cluster trains vision, time-series, and satellite models. Federated learning improves regional models while raw production data never leaves the farm. Alerts and feeding or aeration advice return to the pond, including offline-first operation for sea cages.

Platform layers

AquaSense

Field layer · farms and cooperatives
Water quality, equipment, cameras, and real-time alarms at the pond or cage.

AquaBrain

Decision layer · groups and processors
Disease warning, precision feeding, yield and water-quality forecast.

BlueRisk

Risk layer · insurers, exporters, government
Climate, disease, regional environment, GAP/ASC evidence, and supply-chain risk.

What the stack does

Marine climate early warning from satellite and buoy data: storm, rain, salinity shock, and algal crash risk
24–72 hour forecasts for dissolved oxygen, pH, salinity, and ammonia, moving from alarms to prediction
Underwater vision for fish behavior, surface lesions, and feeding activity to cut FCR
Federated compute: A100 center, district Orin hubs, pond terminals; raw data stays on site

Technology

Built for production inference, not slideware.

Edge Hub inferenceNVIDIA A100 training clusterNVIDIA Jetson Orin at the farm and districtUnderwater computer vision24–72h water-quality and climate forecastTHEOS-2 satellite and marine buoy fusionFederated learningZero-trust and PDPA-aligned data controls